Académique Documents
Professionnel Documents
Culture Documents
Abstract
Fund selection is an important issue for investors. This topic has spawned abundant
academic literature. Nonetheless, most of the time, these works concern only active
management, whereas many investors, such as institutional investors, prefer to invest in
index funds. The tools developed in the case of active management are also not suitable
for evaluating the performance of these index funds. This explains why information
ratios are usually used to compare the performance of passive funds. However, we
show that this measure is not pertinent, especially when the tracking error volatility of
the index fund is small. The objective of an exchange traded fund (ETF) is precisely
to offer an investment vehicle that presents a very low tracking error compared to its
benchmark. In this paper, we propose a performance measure based on the value-at-risk
framework, which is perfectly adapted to passive management and ETFs. Depending
on three parameters (performance difference, tracking error volatility and liquidity
spread), this efficiency measure is easy to compute and may help investors in their
fund selection process. We provide some examples, and show how liquidity is more of
an issue for institutional investors than retail investors.
Keywords: Passive management, index fund, ETF, information ratio, tracking error, liq-
uidity, spread, value-at-risk.
JEL classification: G11.
1 Introduction
The market portfolio concept has a long history and dates back to the seminal work of
Markowitz (1952). In that paper, Markowitz defines precisely what portfolio selection means:
“the investor does (or should) consider expected return a desirable thing and variance of re-
turn an undesirable thing”. Indeed, Markowitz shows that an efficient portfolio is a portfolio
that maximizes the expected return for a given level of risk (corresponding to the variance
of return). Markowitz concludes that there is not only one optimal portfolio, but a set of
optimal portfolios called the efficient frontier (represented by the solid blue curve in Figure
1). By studying liquidity preference, Tobin (1958) shows that the efficient frontier becomes a
∗ We are profoundly grateful to Bou Ly Wu for his support with data management and the computation
of liquidity spreads on limit order books. We would also like to thank Arnaud Llinas, Raphaël Dieterlen,
Valérie Lalonde, François Millet and Matthieu Mouly for their helpful comments.
1
Measuring Performance of Exchange Traded Funds
One of the difficulties faced when computing the tangency portfolio is that of precisely
defining the vector of expected returns of the risky assets and the corresponding covari-
ance matrix of returns. In 1964, Sharpe developed the CAPM theory and highlighted the
relationship between the risk premium of the asset (the difference between the expected
return and the risk-free rate) and its beta (the systematic risk with respect to the tangency
portfolio). Assuming that the market is at equilibrium, he showed that the prices of assets
are such that the tangency portfolio is the market portfolio, which is composed of all risky
assets in proportion to their market capitalization. That is why we use the terms, tangency
portfolio and market portfolio indiscriminately nowadays.
Another step forward was made by Jensen (1968), who measured the performance of 115
(equity) mutual funds using the alpha measure and concluded that:
“The evidence on mutual fund performance indicates not only that these 115
mutual funds were on average not able to predict security prices well enough to
2
Measuring Performance of Exchange Traded Funds
The seminal work of Michael Jensen is the starting point of the development of passive
management. Indeed, the first index fund was launched in 1971 by John McQuown at Wells
Fargo Bank for the Samsonite luggage company (Bernstein, 1992). It took a long time for
investors to accept the ideas of Markowitz, Sharpe and Jensen, but passive management
represents now a large part of the asset management industry as shown in Figure 2.
16%
14%
12%
10%
8%
6%
4%
2%
0%
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
The paper is organized as follows. Section 2 presents the efficiency measure, which is in
fact a risk measure. In particular, we describe the reasoning behind this indicator and its
rationale. Empirical results are reported in Section 3. In section 4, we consider different
variations on the tracker efficiency measure. For example, we consider other definitions of
the risk measure and the liquidity spread. Finally, in Section 5 we draw our conclusions
3
Measuring Performance of Exchange Traded Funds
• Picking one or more elements of this universe by combining quantitative and qualitative
criteria.
The first step is crucial in order to define a universe of mutual funds, that are sufficiently
homogeneous in terms of risk analysis. For instance, if we want to invest in equities, it
could be useful to describe the universe more accurately. Does the investment concern
global equities, regional equities or country equities? Does the investment definition relate to
specific sectors, by focusing on or excluding some of them? What is the bias of the investment
universe in terms of style analysis? The capacity of the investor to clearly define the borders
of the universe is the most important part of the fund picking process, because the second
step is relatively straightforward once the first step has been accurately completed. For
example, investing in large capitalization equities located in the Eurozone that do not belong
to the banking system, and are recognized as socially responsible, effectively reduces the
universe to a small number of mutual funds.
Nevertheless, defining the investment universe is not always straightforward, for many
reasons. First, the category may be very large with numerous mutual funds. Suppose we
wanted to invest in sovereign bonds in the Eurozone. According to the Morningstar database
there are more than 200 such funds available. Second, a category may not necessarily exist
if the criteria are too specific1 . Third, investment styles are heterogeneous, because they
cannot be defined in a precise way. Take value and growth styles, for example. Everybody
knows what they mean, but if we ask two people to classify the components of the S&P 500
index according to growth/value risk factors, we will get two different answers. This shows
that investing in active management may have a significant cost, because thorough research
is necessary to find accurate information. That is why, in the end, many investments are
based on past performance.
ing German bonds (because the yield is too low) and Spanish bonds (because they are too risky), would
struggle to find mutual funds that satisfy these criteria.
4
Measuring Performance of Exchange Traded Funds
' $
The ratings are constructed by combining three criteria. First, one measures extreme
risks using a Cornish-Fisher value-at-risk at a 99% confidence level. If the VaR of the
fund is too high, the fund is not rated. Second, the alpha of the fund is estimated. It is
done in two steps. In a first step, the style analysis of Sharpe is considered to build the
composite benchmark of the fund. When this benchmark is selected, one deduces alpha
by a regression analysis:
∑
n ( )
(j)
Rt − r = α + βj Rt − r + εt
j=1
(j)
where Rt is the return of the fund, Rt is the return of the ith selected benchmark and
r is the risk-free rate. The alpha criterion allows us to distinguish funds rated below and
above 3 stars. For example, the 4 and 5 stars correspond to funds which have a strictly
positive alpha. The distinction between 4 and 5 stars is done using a gain frequency
measure. It corresponds to the number of times in percent that the fund has delivered
performance that was better than that of its benchmark. If the gain frequency measure
is less than 50%, the fund is rated 4 stars, otherwise it is rated 5 stars. Finally, a super
rating is attributed to funds rated 5 stars and which have a Hurst exponent H larger
than 1/2.
& %
The first factor is important because it impacts on the net performance of the invest-
ment vehicle. For instance, high management fees generally have a negative impact on the
performance of mutual funds. Indeed, Elton et al. (1993) investigated the informational
efficiency of mutual fund performance for the period 1964-1984 and concluded:
“There is statistically significant relationship between alpha and expenses and
a relationship that can clearly be seen by examining the quintiles. Higher ex-
penses are associated with poorer performance. Management does not increase
performance by an amount sufficient that justify higher fees”.
The results are the same for index funds (Elton et al., 2004). Nevertheless, management
fees are not the only factor that impact directly on the return of an index fund. We can cite
for example brokerage costs, dividend taxes, etc. A fund may also benefit from revenues
derived from securities lending, fiscal optimization or index arbitrage2 .
The second factor measures the relative risk of the fund with respect to the index. It
corresponds in general to the volatility of the tracking error, which is defined as the difference
2 Index arbitrage concerns addition/deletion mechanisms of the index and corporate actions as subscrip-
5
Measuring Performance of Exchange Traded Funds
between fund and index returns. It indicates how the tracker performance may deviate from
the index performance. In the words of Harry Markowitz, it is ‘an undesirable thing’ for
the investor. This performance difference comes from the fact that replication strategies
cannot be perfect, because the invested portfolio can not track the index weights exactly
all the time. We generally distinguish between three replication methods: full replication,
optimized sampling and synthetic replication. In full replication, the fund manager invests in
all index securities in the same proportion as the underlying index. But the portfolio weights
can never be identical to the theoretical index weights, unless the fund manager buys the
whole market3 . Moreover, the portfolio weights are not static, as a result of subscriptions
and redemptions, dividends, cash drag, etc.
The third factor is a key element for passive funds. The liquidity of the fund is inherently
connected to the liquidity of the index. A S&P 500 index fund is certainly more liquid than
an MSCI emerging markets index fund. But the investment universe is not the only factor
in fund liquidity. These funds are mainly bought by institutional investors, especially in
Europe. Liquidity is then crucial when they carry out subscriptions or redemptions. These
decisions may have a big impact on the market. Moreover, the fund manager has little
margin for discretionary decisions, because he has to follow the systematic index strategy.
The index ignores more and less liquidity, but the index fund does not. Some recent events,
like the March 2011 Tohoku earthquake in Japan or the Greek debt crisis, have shown
how difficult it can be to manage the liquidity of an index fund in the face of such stress
scenarios. Redemptions from Japanese equity index funds were difficult to execute in the
days after the catastrophe in northern Japan on Friday March 11, 2011. In the same way,
the decision of EuroMTS to keep Greek bonds in the EuroMTS Eurozone Govies index (ex
CNO) after Greece was downgraded from investment grade in April 2010 caused problems in
managing index funds that were benchmarked to this index. The problem is more complex
with trackers, because they offer intra-day liquidity. And this liquidity has a cost, which
can be measured by bid-ask spreads.
Another element that is important for investors is the operational structure of an ETF.
We generally distinguish two main structures (physical and synthetic) used in the European
markets. Even if the operational structure may be sometimes an important criteria for
investors, it is more a go/no go check test than a statistic that can be taken into account in
the design of an efficiency measure.
For example, if N1,t = 136017, N2,t = 87123, P1,t = 110.54, N2,t = 125.23, we have w1,t = 57.95% and
w2,t = 42.05%. If the value of the fund is $1000, full replication involves buying respectively 5 and 3 shares
of the first and second stocks, whereas the cash represents $71.61. The portfolio weights of the two stocks
are then w1,t = 55.27% and w2,t = 37.57%. At time t + 1, we assume that there is a subscription of $60. If
the prices remain the same, we could buy another share of either the first stock or the second stock. In the
two cases, we obtain portfolio weights that are different from the index weights.
6
Measuring Performance of Exchange Traded Funds
the asset i. The tracking error of the active portfolio x with respect to the benchmark is the
difference between the return of the portfolio R (x) and the return of the benchmark R (b):
e = R (x) − R (b)
∑n ∑
n
= xi Ri − bi Ri
i=1 i=1
⊤ ⊤
= x R−b R
⊤
= (x − b) R
The expected tracking error is defined as follows:
µ (x | b) = E [e]
⊤
= (x − b) µ
whereas tracking error volatility is equal to4 :
σ 2 (x | b) = σ 2 (e)
⊤
= (x − b) Σ (x − b)
The objective of the investor is to maximize the expected tracking error with a constraint
on the tracking error volatility:
⊤
x⋆ = arg max (x − b) µ
u.c. 1⊤ x = 1 and σ (e) ≤ σ ⋆
We gave an example of the tracking-error efficient frontier in Figure 3. It is a straight line
when there is no restriction (Roll, 1992). If we impose some constraints, the efficient frontier
is moved to the left and is no longer a straight line.
To compare the performance of different portfolios, a better measure than the Sharpe
ratio is the information ratio, which is defined as follows (Grinold and Kahn, 2000):
µ (x | b)
IR (x | b) =
σ (x | b)
⊤
(x − b) µ
= √
⊤
(x − b) Σ (x − b)
If we consider a combination of the benchmark b and the active portfolio x, the composition
of the portfolio is:
y = (1 − α) b + αx
with α ≥ 0 the proportion of wealth invested in the portfolio x. We get:
⊤
µ (y | b) = (y − b) µ = αµ (x | b)
and:
⊤
σ 2 (y | b) = (y − b) Σ (y − b) = α2 σ 2 (x | b)
We deduce that:
µ (y | b) = IR (x | b) · σ (y | b)
It is the equation of a linear function between the expected tracking error and the tracking
error volatility of the portfolio y. It implies that the efficient frontier is a straight line:
4 In IOSCO terminology, µ (x | b) corresponds to Tracking Difference (TD) whereas σ (x | b) is called
7
Measuring Performance of Exchange Traded Funds
If we add some other constraints to the portfolio optimization problem, the efficient frontier
is no more a straight line. In this case, one optimized portfolio dominates all the other
portfolios. It is the portfolio that belongs to the efficient frontier and the straight line that
is tangent to the efficient frontier. It is also the portfolio that maximizes the information
ratio. An illustration is provided in Figure 4. In this case, we check that the tangency
portfolio is the one that maximizes the tangent θ, which is the information ratio:
BC
tan θ =
AB
µ (x | b)
=
σ (x | b)
= IR (x | b)
8
Measuring Performance of Exchange Traded Funds
Proposition 1 Let x and y be two portfolios benchmarked on the same index b. x is prefer-
able to y if and only if the information ratio of x is greater than the information ratio of
y:
x ≻ y ⇔ IR (x | b) ≥ IR (y | b)
y = (1 − α) x0 + αx
5 We assume that the information ratio is positive.
9
Measuring Performance of Exchange Traded Funds
Let us illustrate this drawback with an example. We assume that µ (x0 | b) = −5 bps,
µ (x1 | b) = 45 bps, σ (x1 | b) = 45 bps, σ (x1 | x0 ) = 25 bps and σ (x0 | b) = 5 bps. Portfolio
x1 and tracker x0 are represented in Figure 6. Next we consider a second portfolio x2 with the
following characteristics µ (x2 | b) = 15 bps and σ (x2 | b) = 15 bps. We get IR (x1 | b) = 1.29
and IR (x2 | b) = 1.0. Using Proposition 1, we can deduce that x1 is better than x2 because
we could build the portfolio x3 that dominates x2 :
x3 ≻ x2 ⇒ x1 ≻ x2
But the problem is that the information ratio of x3 could not reached because the benchmark
could only be replicated by tracker x0 . In this case, the combination of x1 and x0 gives x4
6 Due to the management fees.
10
Measuring Performance of Exchange Traded Funds
This problem concerns benchmarked funds with low tracking-error volatility in particular.
In this case, the information ratio cannot be used to select funds. This is especially true
for trackers that aim to have the lowest tracking-error volatility. Moreover, the information
ratio is not appropriate for measuring the relative performance of the tracker that presents
the lowest tracking error. Indeed, it is generally a fund that has a small but negative excess-
return performance. Because its information ratio is negative, this tracker is not efficient
from the information ratio perspective. It means that no investors are interested in using
it. This theoretical point of view is in contradiction with practice, because passive investors
like to consider funds with the smallest tracking error volatility.
Remark 1 Another problem arises from the fact that the information ratio ignores the
magnitude of tracking-error volatility. For instance, if we consider two trackers with the
following characteristics:
the second tracker is preferred to the first tracker if we use the information ratio as the
selection criteria. However, the first tracker does a better replication job than the second
one.
11
Measuring Performance of Exchange Traded Funds
In Figure 8, we show the impact of the different factors on the efficiency measure. The
solid line corresponds to the previous parameter values whereby we modify one parameter
value for each dashed line. For instance, in the first panel, we consider that µ (x | b) is
equal to 70 bps. This tracker is preferred as the original tracker because it has a better
excess-return performance, the same tracking error volatility and the same bid-ask spread.
In the second panel, the second tracker has a wider spread8 . In this case, we prefer the
first tracker. For the third panel, we obtain the same conclusion if the second tracker has a
greater tracking error volatility. We can then deduce the following result:
7 We consider the opposite of the loss in order to obtain an ascending order: the bigger the efficiency
12
Measuring Performance of Exchange Traded Funds
13
Measuring Performance of Exchange Traded Funds
Proposition 2 Let x and y be two trackers benchmarked to the same index b. x is preferable
to y if and only if the efficiency measure of x is larger than the efficiency measure of y:
x ≻ y ⇔ ζα (x | b) ≥ ζα (y | b)
ζα (x | b) = 40 − 20 − 1.65 × 30
= −29.5
and:
ζα (y | b) = 30 − 15 − 1.65 × 20
= −18.0
Because ζα (y | b) > ζα (x | b), we can deduce that tracker y is more efficient than tracker x.
A loss of more than 29.5 bps has a 5% probability of occurring for tracker x. In the case of
tracker y, this loss is only 18 bps for the same level of probability.
3 Empirical results
In this paragraph, we apply the efficiency measure ζα (x | b) to the European ETF market.
The confidence level is set at 95%. Let [t0 , T ] be the study period with n trading dates. We
compute ζα (x | b) as follows:
ζα (x | b) = µ̂ (x | b) − ŝ (x | b) − 1.65 · σ̂ (x | b)
where µ̂ (x | b), ŝ (x | b) and σ̂ (x | b) are the estimated statistics for the given study period.
We have:
( )1/(T −t0 ) ( )1/(T −t0 )
VT (x) VT (b)
µ̂ (x | b) = −
Vt0 (x) Vt0 (b)
1 ∑
T
ŝ (x | b) = st (x | b)
n t=t
0
1 ∑T
1 ∑ T
µ̄ (x | b) = Rt (x) − Rt (b)
n − 1 t=t +1 n − 1 t=t +1
v
0 0
u
u 260 ∑ T
= t
2
σ̂ (x | b) (Rt (x) − Rt (b) − µ̄ (x | b))
n − 1 t=t +1
0
where Vt (x) (or Vt (b)) is the net asset value of the tracker (or benchmark) at time t, st (x | b)
is the bid-ask spread9 at time t and Rt (x) (or Rt (b)) is the daily return of the tracker (or
the benchmark) defined as follows:
Vt (x)
Rt (x) = −1
Vt−1 (x)
9 More precisely, we compute the best spread of the first limit order for each listing place and each trading
day t. st (x | b) is then the weighted average by considering the daily volume of the different listing places.
14
Measuring Performance of Exchange Traded Funds
Results10 are reported in Tables 1, 2 and 3. According to the efficiency measure ζα (x | b),
the best tracker for the Eurostoxx 50 index is db X-trackers, followed by Amundi and Lyxor.
We notice that ζα (x | b) assigns a positive value to most of the trackers, due probably to
10 All the statistics are expressed in bps except µ̂ (x) and σ̂ (x) which are measured in %.
15
Measuring Performance of Exchange Traded Funds
fiscal optimization of dividends. These results are different for the S&P 500 and MSCI World
indices. Indeed, ζα (x | b) is generally negative. It is due to lower performance and a wider
spread, even if ETFs tracking the S&P 500 and MSCI World indices have less tracking-error
volatility11 than for the Eurostoxx 50 index. At the end, Lyxor is ranked top, followed by
HSBC and Credit Suisse for the S&P 500 index, and Commerzbank and Amundi for the
MSCI World index.
For instance, in our example, the reference index is composed by 49.50% of the Topix
index and 50.50% of the MSCI Japan index12 , and we obtain the results given in Table 5.
For Japanese equities, iShares appears then the best tracker.
11 Nevertheless, we note some divergence in the case of MSCI World trackers. This is even more so when
16
Measuring Performance of Exchange Traded Funds
We notice that the ranking changes slightly in comparison to the previous one. Amundi is
thus the best tracker, followed by db X-trackers and Lyxor.
Looking at the results in Table 7, the ranking of S&P 500 trackers is unchanged: Lyxor
remains the best tracker, followed by HSBC and Credit Suisse. We conclude that the choice
of semi-variance over volatility does not fundamentally alter the results. But the efficiency
measure based on semi-variance has an important drawback. It is less sensitive to tracking
errors, which is however an important criteria for investors. Another important point is that
semi-variance is just another way of measuring the skewness of a distribution:
17
Measuring Performance of Exchange Traded Funds
If we compute this ratio for the previous trackers, we get a range between 1.9 and 14.8
for the Eurostoxx 50 index and between 1.9 and 6.6 for the S&P 500 index. This clearly
demonstrates that tracking errors exhibit high skewness. This result is confirmed by the
Kernel analysis of empirical density (see Figures 9, 10 and 11). In this context, we think
that replacing volatility by semi-volatility is not a good method. It would be better to adopt
another risk measure that takes the skewness of tracking errors into account.
18
Measuring Performance of Exchange Traded Funds
19
Measuring Performance of Exchange Traded Funds
In Tables 8 and 9, we compute the efficiency indicator ζα (x | b) with five risk measures:
σ (x | b) corresponds to volatility, σ− (x | b) is below-mean semi-volatility, VaRα (x | b) is
historical value-at-risk, ESα (x | b) corresponds to the historical expected shortfall, whereas
CFα (x | b) uses the Cornish-Fisher expansion to compute value-at-risk. In the case of Eu-
rostoxx 50 trackers, we observe some minor differences. Indeed, Lyxor turns out to be the
best tracker if we are looking at historical value-at-risk. The rank of db X-trackers varies
between 1 and 3 depending on the risk measures. For S&P 500 trackers, we observe a good
consistency across the various rankings.
20
Measuring Performance of Exchange Traded Funds
In Figure 12, we shown the change in the spread sN (x | b) for the Amundi Eurostoxx
50 tracker and different values of the notional N . We notice that this spread changes with
respect to market liquidity. For each tracker and each notional value, we have shown the
corresponding boxplot in Figure 13. The boxplot indicates the minimum value, the quartile
range, the median and the last decile. We check that:
N1 ≥ N2 ⇒ sN1 (x | b) ≥ sN2 (x | b)
We notice also that for some trackers, the spread increases greatly in line with volume.
In Figure 14, we have drawn some scatterplots. It is interesting to note that when liquidity
is not an issue, the trackers are more or less equivalent in terms of spread14 . This is not the
case when there are some liquidity problems (see Figure 15).
13 Computational details are provided in Appendix A.5.
14 If we consider two trackers x and y, we have:
sN (x | b) ≃ sN (y | b)
if sN (x | b) ≃ 0 and sN (x | b) ≃ 0.
21
Measuring Performance of Exchange Traded Funds
22
Measuring Performance of Exchange Traded Funds
Let us now compute the efficiency indicator ζα⋆ (x | b) by considering this new spread
measure. The results are shown in Table 10. We consider two calculations: one based on
the median spread and another one based on the quantile at the 95% confidence level. We
notice that this new measure has a big impact on the efficiency indicator. For instance, if
we consider a notional of 2 MEUR, we obtain a different ranking.
Table 10: Impact of the liquidity on the efficiency measure (Eurostoxx 50)
Remark 2 In our definition of the efficiency indicator, we assume that there is one trade
each year. If there are m trades per year, the performance measure becomes:
ζα (x | b) = µ (x | b) − m · s (x | b) − Φ−1 (α) σ (x | b)
This formula highlights the importance of liquidity for active managers. For instance, a
highly active manager will only be interested in the spread measure because:
lim ζα (x | b) = −m · s (x | b)
m→∞
23
Measuring Performance of Exchange Traded Funds
5 Conclusion
In this paper, we have developed a performance measure to compare passive management,
in particular tracker investment vehicles such as exchange traded funds. This measure is
very different from the ones used to assess the performance of active management. It is a
value-at-risk measure based on three parameters: the performance difference between the
fund and the index, the volatility of the tracking error and the liquidity spread. This simple
measure may be easily implemented by investors or rating agencies in relation to the fund
picking process.
This paper also highlights the role of liquidity spread in measuring the efficiency of an
ETF. It is particular true for institutional investors, who may subscribe or redeem large
investment amounts. The liquidity spread is also the most important parameter for active
managers who use ETFs to implement their convictions in line with their tactical asset
allocation.
24
Measuring Performance of Exchange Traded Funds
References
[1] Amenc N. and Martellini L. (2002), The Brave New World of Hedge Fund Indices,
EDHEC-Risk Working paper.
[2] Barras L., Scaillet O. and Wermers R. (2010), False Discoveries in Mutual Fund
Performance: Measuring Luck in Estimated Alphas, Journal of Finance, 65(1), pp.
179-216.
[3] Bawa V.S. (1975), Optimal Rules for Ordering Uncertain Prospects, Journal of Finan-
cial Economics, 2(1), pp. 95-121.
[4] Beasley J.E., Meade N. and Chang T.-J. (2003), An Evolutionary Heuristic for
the Index Tracking Problem, European Journal of Operational Research, 148(3), pp.
621-643.
[5] Blake C.R., Elton E.J. and Gruber M.J. (1993), The Performance of Bond Mutual
Funds, Journal of Business, 66(3), pp. 371-403.
[6] Bernstein P.L. (1992), Capital Ideas: The Improbable Origins of Modern Wall Street,
Free Press.
[7] Bernstein P.L. (2007), Capital Ideas Evolving, John Wiley & Sons.
[8] Carhart M.M. (1997), On Persistence in Mutual Fund Performance, Journal of Fi-
nance, 52(1), pp. 57-82.
[9] Dieterlen R. and Hereil P. (2012), Index Arbitrage Explained, Journal of Indexes
Europe, 2(6), pp. 18-23.
[10] Elton E.J., Gruber M.J., Das S. and Hlavka M. (1993), Efficiency with Costly
Information: A Reinterpretation of Evidence from Managed Portfolios, Review of Fi-
nancial Studies, 6(1), pp. 1-22.
[11] Elton E.J., Gruber M.J. and Busse J.A. (2004), Are Investors Rational? Choices
among Index Funds, Journal of Finance, 59(1), pp. 261-288.
[12] Estrada J. (2007), Mean-semivariance Behavior: Downside Risk and Capital Asset
Pricing, International Review of Economics and Finance, 16(2), pp. 169-185.
[13] Fishburn P.C. (1977), Mean-Risk Analysis with Risk Associated with Below-Target
Returns, American Economic Review, 67(1), pp. 116-126.
[14] Gastineau G.L. (2002), Equity Index Funds Have Lost Their Way, Journal of Portfolio
Management, 28(2), pp. 55-64.
[15] Gastineau G.L. (2004), The Benchmark Index ETF Performance Problem, Journal
of Portfolio Management, 30(2), pp. 96-103.
[16] Grinblatt M. and Titman S. (1989), Portfolio Performance Evaluation: Old Issues
and New Insights, Review of Financial Studies, 2(3), pp. 393-421.
[17] Grinold R.C. and Kahn R.N. (2000), Active Portfolio Management: A Quantita-
tive Approach for Providing Superior Returns and Controlling Risk, Second edition,
McGraw-Hill.
25
Measuring Performance of Exchange Traded Funds
[24] Nawrocki D.N. (1999), A Brief History of Downside Risk Measures, Journal of In-
vesting, 8(3), pp. 9-25.
[25] Pope P.F. and Yadav P.K. (1994), Discovering Errors in Tracking Error, Journal of
Portfolio Management, 20(2), pp. 27-32.
[26] Prigent J.-L. (2007), Portfolio Optimization and Performance Analysis, Chapman &
Hall.
[28] Sharpe W.F. (1964), Capital Asset Prices: A Theory of Market Equilibrium under
Conditions of Risk, Journal of Finance, 19(3), pp. 425-442.
[29] Tobin J. (1958), Liquidity Preference as Behavior Towards Risk, Review of Economic
Studies, 25(2), pp. 65-86.
26
Measuring Performance of Exchange Traded Funds
A Technical appendix
A.1 Proof of the equation (1)
We have:
⊤
σ 2 (x | b) = (x − b) Σ (x − b)
= x⊤ Σx + b⊤ Σb − 2x⊤ Σb
= σ 2 (x) + σ 2 (b) − 2ρ (x, b) σ (x) σ (b)
We deduce that the correlation between the portfolio x and the benchmark b is:
σ 2 (x) + σ 2 (b) − σ 2 (x | b)
ρ (x, b) =
2σ (x) σ (b)
It follows that:
⊤ ( )
2 (x0 − b) Σ (x − b) = 2 x⊤ ⊤ ⊤ ⊤
0 Σx − x0 Σb − x Σb + b Σb
= σ 2 (x0 ) + σ 2 (x) − σ 2 (x | x0 ) −
σ 2 (x0 ) − σ 2 (b) + σ 2 (x0 | b) −
σ 2 (x) − σ 2 (b) + 2σ 2 (x | b) + σ 2 (b)
= σ 2 (x0 | b) + σ 2 (x | b) − σ 2 (x | x0 )
y = (1 − α) x0 + αx
We deduce that:
y − b = (1 − α) (x0 − b) + α (x − b)
It follows that:
µ (y | b) = (1 − α) µ (x0 | b) + αµ (x | b)
and:
⊤
σ 2 (y | b) = (y − b) Σ (y − b)
2 ⊤
= (1 − α) (x0 − b) Σ (x0 − b) +
⊤
α2 (x − b) Σ (x − b) +
⊤
2α (1 − α) (x0 − b) Σ (x − b)
2
= (1 − α) σ 2 (x0 | b) + α2 σ 2 (x | b) +
( )
α (1 − α) σ 2 (x0 | b) + σ 2 (x | b) − σ 2 (x | x0 )
( )
= (1 − α) σ 2 (x0 | b) + ασ 2 (x | b) + α2 − α σ 2 (x | x0 )
We deduce that:
µ (y | b)
IR (y | b) =
σ (y | b)
(1 − α) µ (x0 | b) + αµ (x | b)
= √
(1 − α) σ (x0 | b) + ασ 2 (x | b) + (α2 − α) σ 2 (x | x0 )
2
27
Measuring Performance of Exchange Traded Funds
We have:
Pr {L (x | b) ≤ ζ} = α
⇔ Pr {e − s ≤ ζ} = α
⇔ Pr {e ≤ s + ζ} = α
{ }
e − µ (x | b) s + ζ − µ (x | b)
⇔ Pr ≤ =α
σ (x | b) σ (x | b)
( )
s + ζ − µ (x | b)
⇔ Φ =α
σ (x | b)
It follows that:
s + ζ − µ (x | b)
= Φ−1 (α)
σ (x | b)
or:
ζ = s − µ (x | b) + Φ−1 (α) σ (x | b)
We deduce that:
ζα (x | b) = −ζ
= µ (x | b) − s − Φ−1 (α) σ (x | b)
Ω = V ΛV ⊤ (3)
where V is the matrix of the eigenvectors and Λ = diag (λ1 , . . . , λm ) is a diagonal matrix
of eigenvalues15 . Because the decomposition (3) corresponds to the principal component
analysis (PCA), we may extract the main representative factor defined by:
∑
m
Ft = Vj,1 Rt (bj )
j=1
As noticed by Amenc and Martellini (2002), this first component of the PCA maximizes the
representativeness of the set of competing indices. The reference index is then a weighted
portfolio of the competing indices:
∑m
b= wj bj
j=1
28
Measuring Performance of Exchange Traded Funds
µ = E [X]
∫ ∞
= x dF (x)
−∞
∫ ∞
= xf (x) dx
−∞
The variance of X corresponds to the second moment: σ 2 (X) = µ2 (X). σ (X) is called
standard deviation, but it is well known as volatility in finance, where X represents the
return on an asset.
29
Measuring Performance of Exchange Traded Funds
where sj is the spread of the j th tick and tj+1 − tj the elapsed time between two consecutive
ticks: ( + )
P̄j − P̄j−
sj = cj
P̄j0
We have also: ∑K
Q̄•j,k Pj,k
•
P̄j•
k=1
= ∑K
k=1 Q̄•j,k
+ −
where Pj,k (resp. Pj,k ) is the ask (or bid) price at tj for the k th limit order. The average
0
mid price P̄j corresponds to:
P̄j+ + P̄j−
P̄j0 =
2
−
The quantity Q̄+
j,k and Q̄j,k are defined as follows:
( ( ))
∑
k−1
Q̄•j,k = max 0, min Q•j,k , Q⋆j − Q•j,l
l=1
−
Here, Q+j,k and Qj,k are the ask and bid volumes of the k
th
limit order. The reference
⋆
quantity Qj is the ratio between the trading notional N and the mid price:
N
Q⋆j =
P̄j0
Sometimes it may appear that the trading volume on the order book is lower than the
notional N . That is why the factor cj may be greater than one:
Q⋆j
cj = max 1, (∑ )
K + ∑K −
min k=1 Qj,k , k=1 Qj,k
For instance, if we wish to execute an order of 2 MEUR and there is only a trading volume
of 1 MEUR, we multiply the spread by two.
Remark 3 For each trading day, we compute the daily spread for the different listing places
using the previous formulas and we take the best spread.
A.5.2 Example
Let us illustrate the spread calculation with the order book given in Table 11. For Q⋆ = 1000,
we get the results shown in Table 12. We deduce that the mid price P̄j0 is 85.98 whereas the
spread sj is equal to 20.12 bps. It corresponds to a notional N = Q⋆ × P̄j0 of 85 981 e.
30
Measuring Performance of Exchange Traded Funds
If we wish to execute an order for 100 KEUR, it implies that Q⋆ > 1000. Given a notional
N , it is then possible to find the optimal value of Q⋆ :
{ }
Q⋆ = inf Q ∈ N : QP̄j0 ≥ N
We may solve this nonlinear inequality by using a bisection algorithm. For instance, if N is
equal to 100 KEUR, we obtain Q⋆ = 1163. In this case, the spread is equal to 23.24 bps.
Sometimes, there are not enough orders in the book. For instance, if N = 500 KEUR, the
mid price is 85.972 meaning that Q⋆ = 5816. The coefficient c is then larger than 1 and we
obtain a spread equal to 87.81 bps.
31
Measuring Performance of Exchange Traded Funds
32